Everyone assumes the same thing: rank in the top 10, and the AI answers will cite you.
It’s the comforting story the industry has been telling itself since AI Overviews launched — that AI visibility is just organic visibility wearing a new interface. Keep doing SEO; the citations will follow.
- We tested it. With SERP data captured from Google, against Google's own rankings.
The result:
For the same query, on the same engine, on the same day the pages Google’s AI cites and the pages Google ranks are mostly different.
- But here's the part fewer people will quote, and it's in the same dataset:
Ranked pages were still cited at a rate no other source category comes close to.
The most commercial queries had no AI Overview at all
— the classic click economy fully intact.
On settled, single-answer queries, the rankings and the citations were identical.
So this isn’t a eulogy for SEO. It’s a map of where ranking is still the whole game, and where it’s now only the entry ticket.
Key Findings (TL;DR)
86%
of SERPs carried an AI Overview (across 14 education-sector queries, US).
35.5%
pointed to a URL ranking in Google's top 10 for that same query.
18.3%
came from the top 3 — the positions the entire industry fights over.
52.7%
pointed to pages whose domains don't appear anywhere in the top 10. Not a lower position on the same site. Completely absent.
11.8%
of cases, the domain ranked — but Google's AI cited a different page from that domain than the one ranking.
- The strongest pattern in the data
the bigger the AI Overview, the less it resembles the rankings.
- Compact 3-citation Overviews drew 100% from top-ranking URLs
- The largest Overview (27 sources) cited only 2 ranking pages
- The rankings-still-matter column
- Every compact, settled-answer Overview drew exclusively from top-ranking pages
- The most commercial query in the set had no AI Overview at all — the classic click economy fully intact
- Per URL, a top-10 position remained the strongest single predictor of citation — 10 ranked URLs supplied a third of all citations
First, what this data does NOT say
Studies like this get screenshot-quoted, so let’s close the wrong doors before opening the right ones. Three conclusions this data does not support:
It does not say rankings stopped mattering.
Look at the math per page. The ten URLs in the top 10 supplied 35.5% of all citations; the millions of other pages on the web competed for the rest. On a per-URL basis, ranking in the top 10 made a page wildly more likely to be cited than any other position it could occupy.
Ranking is no longer a guarantee of citation — but it remains the single biggest advantage.
It does not say SEO is dead.
Two of our 14 queries had no AI Overview at all — and they were the most commercial ones, the hire-a-service and how-do-I-start searches where businesses actually convert customers. For transactional and commercial intent, the blue links, the local pack, and the click economy are fully intact, and ranking there is still the entire game.
The disruption documented here lives specifically in broad informational queries.
It does not say you can skip SEO and “do GEO” instead.
Even the citations from outside the top 10 weren’t coming from nowhere — many were pages that rank for adjacent queries, pulled in when the AI decomposed the head term. Retrieval still runs on the same index, the same crawling, the same link-informed authority signals that SEO builds.
A site invisible to Google’s index is invisible to Google’s AI. The accurate model is layered: ranking infrastructure is the foundation; citation optimisation is a second storey built on top of it. Nobody builds the second storey first.
- What the data does say:
- on broad informational queries, the foundation alone no longer finishes the building. Hold both of those at once, and the rest of the findings make sense.
Why this question matters now
Two narratives are competing for your content budget.
Narrative one
AI search is just search. Citations correlate with rankings, so traditional SEO is GEO, and anyone selling "AI optimisation" is rebranding your existing invoice.
Narrative two
AI retrieval is a different system with different selection criteria, and ranking well is no longer sufficient — or even necessary — to be the source AI quotes.
Almost everything published on this is theory. Opinion pieces, vendor decks, recycled speculation. Very little of it shows actual cited-versus-ranked data for matched queries.
So we pulled the data ourselves.
Methodology
Simple and reproducible:
Query set
14 education-sector queries (study abroad, scholarships, student visas, postgraduate study, education consulting) — a mix of informational and commercial intent.
Market
United States, Google desktop SERPs.
Data source
Ahrefs SERP Overview snapshots, captured June 2026. Each snapshot returns the organic top 10 and the AI Overview’s cited sources for the same SERP, from the same crawl.
Comparison method
For every AI Overview citation, we checked whether:
- The exact URL appeared in the captured top 10 organic results, or
- The URL didn’t, but its domain did, or
- Neither appeared
- Coverage & sample size
Of the 14 SERPs, 12 carried an AI Overview, and 9 of those exposed their full citation list in the snapshot — giving us 93 citations to classify.
One deliberate framing choice
This study measures citations against the top 10 because the top 10 is what SEO strategies, reporting, and client expectations are built on. A cited page ranking #14 is interesting trivia; it’s still invisible in every dashboard that matters.
Finding 1
The majority of citations come from outside the top 10
Let’s start with the number that should reframe your reporting.
52.7% of AI Overview citations — a majority — pointed to pages whose domains appear nowhere in the top 10 for that query.
Add the partial matches, and it gets starker. Barely a third of citations (35.5%) rewarded the exact URLs that earned top-10 positions. The top 3 — where roughly all organic CTR lives, and where every rank-tracking report focuses — accounted for just 18.3% of citations.
- Read that as an exchange rate
For this query set, a top 3 ranking brought roughly a one-in-five chance of being the page Google’s AI actually quoted.
Before anyone panics, hold both halves of this number. A 35.5% share going to just ten URLs means each ranked page is still enormously overrepresented compared to anything else on the web — ranking remains the best citation odds money can buy.
What’s changed is the ceiling: ranking is no longer citation insurance. It’s the strongest ticket in the draw, in a draw that now has other winners.
And here’s the part your dashboard will never volunteer: somewhere in your vertical, a competitor who ranks below you is being quoted above you — inside the answer your customer actually reads. No rank tracker on the market will show you their name.
Which brings us to the pattern that actually explains when the draw opens up.
Finding 2
The bigger the AI Overview, the less the rankings matter
This was the most striking regularity in the dataset, and we haven’t seen it documented elsewhere.
When the AI Overview was compact
- Citing only 3 sources
every single citation came from a top-ranking URL.
Four queries in our set produced these tight, confident Overviews:
"Student visa requirements for Australia"
"Masters in UK for international students"
"What does an education consultant do"
"Cheapest countries to study abroad"
Citation-to-ranking match rate: 12 out of 12
100%
When the AI Overview sprawled
the rankings stopped predicting anything:
“Scholarships for international students”
13 citations, only 3 ranked. 77% from outside.
“How to write a statement of purpose”
18 citations, only 5 ranked. 72% from outside.
- "Study medicine abroad"
27 citations, only 2 ranked in the top 10. 93% from outside.
That last one deserves a pause. Twenty-seven cited sources. The pages holding positions 1 through 10 — earned through years of links and content investment — supplied two of them.
The interpretation is fairly intuitive once you see it. A compact Overview answers a settled question — and for settled questions, the consensus is the top of the SERP. A sprawling Overview is Google’s AI decomposing a broad query into many sub-questions (the “fan-out” behaviour Google has described for its AI surfaces) and retrieving a different source for each fragment: costs from one page, eligibility from another, a first-person account from Reddit, a country comparison from a site you’ve never heard of.
The query you target is no longer the query you’re retrieving for.
You're retrieved for the sub-questions.
Finding 3
Google Cites Your Domain — Just Not The Page You Ranked
The 11.8% middle category is small but operationally important.
In these cases, the domain held a top-10 position — and the AI Overview cited a different URL from the same site.
Real examples from the data:
Ranked
Study-abroad cost guide
Cited
Same site’s scholarship article
Ranked
Medical school directory’s homepage
Cited
Its blog explainer
Ranked
Programme search page
Cited
Editorial listicle from the same domain
- The pattern
Rankings reward the optimised money page; citations reward the page that contains the extractable answer.
Often that’s a supporting article, a blog post, an FAQ — the content many agencies deprioritised as “low value” because it never won the head term.
The pages your content plan stopped updating in 2023 are now doing the job your money page was hired for. Your supporting cluster content just acquired a second job description.
Finding 4
The non-ranking citations aren’t who you’d expect
We classified all 49 citations that came from outside the top 10:
22 (45%)
Commercial and niche sites: consultancies, financial services, ed-tech blogs, regional agencies.
20 (41%)
Institutional sources: universities, government portals, academic libraries, PubMed.
7 (14%)
UGC and social: Reddit threads, YouTube videos, Instagram reels, Facebook posts, Stack Exchange, Medium.
Two things stand out.
- First, authority does not gatekeep citations the way it gatekeeps rankings.
Alongside Oxford and government visa portals, the citation lists included a Bangladeshi education consultancy, a Guyanese university blog, a Caribbean med-school agency, and a Substack-style newsletter — sites that could not crack these top 10s with a decade of link building. They got cited anyway because a passage on their page answered a sub-question cleanly.
A newsletter nobody has heard of out-answered a DR 91 incumbent — and took the citation.
- Second, social content is being cited for queries it doesn't rank for.
An Instagram Reel was cited for a scholarship query. A Facebook post was cited for a medical-study query. These never appear in classic rank tracking — which means classic rank tracking is now blind to a real visibility channel.
Finding 5
Citations borrow from adjacent queries
A subtler pattern, visible once you look at which non-ranking pages get pulled in.
- For "benefits of studying abroad"
Ranked for
N/A (doesn’t rank for benefits)
But cited in the Overview
Major publisher’s “steps to take before studying abroad” article
← Ranks prominently for how to study abroad
- For "cost of studying abroad"
Cited
Student-visa guide
← Page that ranks for an adjacent query
- For "study medicine abroad"
Cited
Residency-pathway explainers
← Pages that own authority for the next-step query
- The AI isn't just re-ranking the SERP in front of it. It's retrieving from the neighbourhood — pages that hold authority for adjacent and follow-up queries get pulled into answers for the head query.
- Notice what this does to the "rankings don't matter" reading
Many of the so-called non-ranking citations are pages that do rank — just for a neighbouring query. Ranking strength didn’t stop feeding citations; it started feeding them sideways. The asset you built to win one keyword is now eligible for citation across the whole topic cluster around it.
The strategic implication
Topical coverage now compounds in a way single-page rankings never did. A cluster that owns the next three questions a searcher would ask gets cited in conversations its individual pages never ranked for.
Inside two SERPs: the extremes of the dataset
Aggregates hide texture, so let’s walk through the two queries at opposite ends of the spectrum.
“Student Visa Requirements For Australia”
Perfect coupling
This is what the comforting narrative looks like when it’s true.
The AI Overview cited exactly three sources
Australian Department of Home Affairs visa page
↓ Ranks at top of organic results
Official Study Australia application guide
↓ Ranks in nearly the same order
Major education publisher’s visa explainer
↓ Same top-order positioning
- Why such tight coupling?
The query has a single, settled, authoritative answer. There is one Subclass 500 visa, one government source of truth, one application process.
There are no sub-questions worth fanning out into, no debate worth synthesising, no firsthand experience that outranks official documentation. When the answer is settled, the consensus and the rankings are the same thing — and the Overview simply quotes the podium.
If your keyword portfolio looks like this — regulated, factual, single-answer — your existing SEO programme is your AI visibility programme. Carry on.
“Study Medicine Abroad”
Total divergence
- The AI Overview for this query cited 27 sources. The top 10 organic results supplied two of them.
Look at what got cited instead, because the variety is the finding:
A pre-med programme directory
A PubMed academic paper on study-abroad outcomes for pre-health students
An Instagram reel
A Facebook news post about UK students moving to Eastern Europe
A Caribbean medical university's blog
A Guyanese university's MBBS article
A Kenyan newspaper feature
A UCAT prep company's UK admissions guide
A residency-pathway explainer
A YouTube video warning about red flags
Alongside Johns Hopkins and NYU
No coherent ranking system produces that list. But a question-decomposition system does.
“Study medicine abroad” silently contains a dozen real questions: Where is it cheapest? Is it respected back home? Can I match into US residency afterwards? Which countries teach in English? What are the red flags?
Each fragment retrieved its own best passage — from whichever corner of the web held it, top 10 be damned.
- Here's the uncomfortable part for anyone running SEO in a competitive vertical
the sites holding those top 10 positions invested years and serious money to get there. For this query, that investment bought a 7% share of the citation surface.
Years of link building. Seven per cent of the answer. Meanwhile, a regional consultancy with no measurable authority got quoted because one paragraph on one page answered one fragment perfectly.
And One More Wrinkle: The Queries With No Overview At All
Two of our 14 queries returned no AI Overview: “study abroad consultant” and the broader “how to study abroad” in its captured snapshot.
The first is the most commercial query in the set — a hire-a-service search with local pack results. Google appears to still treat clearly transactional and local intent as territory where it sends clicks, not answers.
That’s consistent with the broader repositioning argument many of us have been making: informational queries are being absorbed into answers, while commercial and transactional queries remain the durable click economy. This dataset is small, but it points in the same direction.
What This Means For Your SEO Strategy
Let’s translate the data into decisions.
Report citations alongside rankings
Rankings remain the foundation of the report: they capture the commercial queries where clicks still convert, and on settled queries they directly drive citations. But for broad informational queries — the ones AI Overviews dominate — a top 3 position predicted citation less than one time in five, so position improvements alone no longer tell the whole visibility story.
Add citation presence as its own line. The agencies that get hurt by this shift are the ones whose reports go silent on it.
Audit your queries by Overview size
Pull the AI Overview for each priority keyword and count its citations. Three citations? The Overview mirrors the rankings — your existing strategy is your citation strategy. Twelve or more? The Overview is assembling sub-answers from across the index, and you win by owning sub-questions, not by inching from #4 to #2.
Same keyword list, two completely different playbooks.
Re-value your supporting content
The "right domain, wrong page" pattern means the explainers, FAQs, and supporting articles around your money pages are now a retrieval surface in their own right. The pages most agencies labelled "supporting" — and stopped updating in 2023 — are being cited instead of the money pages they support.
Make each one answer its single question completely and extractably: a self-contained passage with the answer up front, specific figures where relevant, and no dependence on surrounding context to make sense.
Build for the fan-out, not just the head term
Map the sub-questions hiding inside each broad target — costs, requirements, timelines, comparisons, risks, "is it worth it," "what goes wrong." Then check whether your cluster contains a direct, dedicated answer to each.
The adjacent-query borrowing in our data means this coverage compounds: a page answering the follow-up question gets cited into the head query's Overview. Topical depth now pays out on queries the individual pages never ranked for.
Small sites: this is the opening
The data shows sites with no realistic shot at these top 10s earning citations on passage quality alone. If you can't outrank the incumbents — and on DR 90 SERPs, you can't — out-answer them on the fragments. One genuinely complete answer to one specific sub-question is now a viable visibility strategy, which has not been true in competitive SEO for a decade.
Track the platforms that rank trackers can't see
Instagram Reels, Facebook posts, and YouTube videos earned citations in this dataset for queries they will never appear in rank tracking for. If your clients' competitors are getting cited from social content while you're only watching blue links, you have a blind spot that won't show up in any tool comparison.
The flip side is an opportunity: it makes social content production a search-visibility channel in its own right, not just community management — a reel or a thread can now hold citation real estate that a webpage can't.
How To Run This Study On Your Own Keywords
The most useful thing about this methodology is that it’s small enough to replicate on any client portfolio in an afternoon. Here’s the process, tool-agnostic:
Step 1 — Pick 10–20 priority queries.
Mix intents deliberately: head informational terms, long-tail questions, and a couple of commercial queries as controls. The contrast between them is where the insight lives.
Step 2 — Capture the SERP and the AI Overview together.
This matters more than it sounds. Comparing today's Overview against last month's rankings introduces noise; you want the citation list and the organic results from the same SERP snapshot. SERP-tracking tools that store AI Overview citations as SERP features (Ahrefs does; several rank trackers now do) give you both in one pull. Manual capture works too — it's just slower.
Step 3 — Classify every citation into three buckets.
Exact URL in the top 10; domain in the top 10 but a different URL cited; neither. A spreadsheet with three columns is genuinely sufficient. Resist the urge to add nuance before you have the basic shape.
Step 4 — Plot non-ranking share against Overview size.
Count of citations on one axis, percentage from outside the top 10 on the other. If your vertical behaves like ours, the line goes up and to the right — and the queries in the top-right corner are where your content strategy needs sub-question coverage, not ranking pushes.
Step 5 — Read the non-ranking citations like a competitor report.
This is the step most people skip and the one with the most leverage. Every citation from outside the top 10 is Google telling you what a ranking page *failed to contain*. A cited Reddit thread means the ranking pages lack firsthand experience. A cited niche explainer means they lack a direct answer to a specific sub-question. That's not abstract insight — it's a content brief.
For agencies, this doubles as a client-facing deliverable. A one-page “cited vs. ranked” snapshot for a client’s top queries communicates the AI search shift more convincingly than any slide of industry statistics — because it’s their keywords.
If You’re Already Investing In SEO: Is That Money Wasted?
No — and this deserves a direct answer, because it’s the question every business owner will ask after reading the findings.
Everything in this study sits on top of a ranking programme, not in place of one. The pages that get cited still have to be crawled, indexed, and trusted — the infrastructure SEO builds. The compact Overviews — and in most B2B verticals, the majority of revenue-driving queries are exactly that kind of specific, settled, commercial query — draw their citations straight from the top of the rankings. The transactional queries where customers actually buy showed no Overview at all, leaving rankings as the entire visibility mechanism. And even in the sprawling Overviews, ranked pages were the single most-cited category, with many “outside” citations earned by pages ranking for neighbouring queries.
So the honest version of the old advice still stands, upgraded: ranking on Google remains the prerequisite for AI visibility — it’s just no longer the finish line for informational queries. The work order hasn’t reversed; it’s extended. Rank first. Then structure the cluster so the AI can quote it.
If your current SEO programme is winning rankings, you haven’t wasted a ringgit, dollar, or pound — you’ve built the first storey. This data just shows there’s now a second one, and which queries need it.
Limitations (Read Before You Quote This)
Honest research states its edges, so here are ours.
This is one vertical (education/study abroad), one market (US), 14 queries, and a single June 2026 snapshot — 93 citations is a meaningful sample for a pattern, not a census. AI Overviews are volatile, and citation sets change between crawls. Our comparison baseline was the captured top ~10 organic results, so a cited page ranking #11–#20 is classified as non-ranking by design — we consider that the right framing (nobody’s strategy celebrates position 14), but it’s a framing choice. And SERP snapshots reflect Ahrefs’ capture of a US desktop SERP; your geography and personalisation will vary.
We’d genuinely like to see this replicated across verticals. If the size-of-Overview pattern holds in finance, health, and B2B SaaS, it’s one of the more useful heuristics in AI-era SEO.
The Final Take: Ranking Is The Entry Fee. The Citation Is The Prize.
The verdict this data supports: the oldest equation in search — #1 ranking = maximum visibility — no longer holds on queries where an AI Overview answers first. More than half the citations in this study came from pages outside the top 10. But the data killed the panic reading too: every overlapping citation traced back to content that ranks somewhere, and domain-level presence remained the strongest single pathway in. Rankings didn’t stop mattering. They stopped being the finish line.
If you're a business owner or marketing lead
Add two questions to your next agency review: "Are we cited in the AI Overviews on our money queries?" and "If not, who is — and why them?" If your reporting can't answer those, you're measuring the race by the old finish line. Keep funding the ranking work — it's the qualifying lap — but expect the content itself to be built for quotation: direct answers, clean structure, citable facts, not just keyword coverage.
If you're an SEO or content specialist
Three changes, starting with the next report: track citations alongside rankings on Overview-bearing queries; restructure key pages so their core answers are extractable (the formats in this study's breakdown — including commercial pages — all got cited); and hunt the gap queries where the domain already ranks but the Overview quotes someone else — those are the cheapest citation wins available, because the entry fee is already paid.
The one-sentence version
Rank to qualify, structure to be quoted, and start reporting the answer box — because the searcher stopped scrolling past it before your dashboard noticed.
FAQ
Does ranking in Google's top 10 get you cited in AI Overviews?
Sometimes — about a third of citations in this study came from top-10 URLs. But the majority (52.7%) came from pages whose domains don’t rank in the top 10 at all, and the share from outside grows sharply as AI Overviews get larger.
Are AI Overview citations and organic rankings pulled from the same system?
They clearly overlap — compact Overviews drew 100% of citations from top-ranking pages in our data. But large Overviews behave like a different retrieval process, decomposing the query into sub-questions and citing passage-level answers from across the index, including pages and platforms that never rank.
Do low-authority sites get cited in AI Overviews?
Yes. Our non-ranking citation pool included small regional consultancies, niche blogs, and newsletter posts alongside Oxford and government portals. Passage relevance appears to outweigh domain authority for citation selection in a way it doesn’t for rankings.
How should agencies track AI visibility?
Separately from rankings: monitor whether the AI Overview exists for each priority query, how many sources it cites, and whether your URLs (or your clients’ competitors’) appear among them — alongside, not instead of, traditional rank tracking.
Is this the same as ChatGPT or Perplexity citation behaviour?
This study covers Google AI Overviews specifically — same engine, same query, same day as the rankings, which is what makes the divergence meaningful. Cross-platform comparison (ChatGPT, Perplexity, Gemini) is the natural follow-up study.
Why do some AI Overviews cite 3 sources and others cite 27?
The pattern in our data suggests Overview size reflects how decomposable the query is. Settled, single-answer questions (visa requirements, role definitions) produce compact Overviews quoting the consensus — which is the top of the SERP. Broad queries containing many implicit sub-questions (costs, comparisons, risks, experiences) produce sprawling Overviews that retrieve a different source per fragment, and that’s where the rankings stop predicting citations.
Does this mean traditional SEO is dead?
No — and the data says so directly. A third of citations still came from top-10 URLs, the compact Overviews drew entirely from top rankings, and the most commercial queries in our set had no Overview at all, leaving the classic click economy intact. The accurate conclusion is narrower and more useful: ranking and being cited are overlapping but distinct outcomes, and on broad informational queries, the overlap is now a minority. Strategies, reporting, and budgets should treat them as two surfaces, not one.
Study data: 14 US education-sector queries, Ahrefs SERP Overview snapshots, June 2026. 93 AI Overview citations classified against captured top-10 organic results. Methodology and full per-keyword table available on request.
